Page 31 - Autonomous Mobile Robots

Visual Guidance for Autonomous Vehicles 15

constraints and the physical degrees of freedom of the camera, yet was stable
enough to work from data on the ground plane alone. As a final note on calib-
ration: any routine should also provide quantified estimates of the uncertainty
of the parameters determined.

1.3 VISUAL GUIDANCE SYSTEMS

1.3.1 Architecture
The modules of a working visual guidance system (VGS) are presented in
Figure 1.2. So far, we have described the key sensors and sensor models. Before
delving into task-specific processes, we need to clarify the role of VGS within
the autonomous vehicle system architecture. Essentially, its role is to capture
raw sensory data and convert it into model representations of the environment
and the vehicle’s state relative to it.

1.3.2 World Model Representation

A world model is a hierarchical representation that combines a variety of sensed
inputs and a priori information [23]. The resolution and scope at each level are
designed to minimize computational resource requirements and to support plan-
ning functions for that level of the control hierarchy. The sensory processing
system that populates the world model fuses inputs from multiple sensors and
extracts feature information, such as terrain elevation, cover, road edges, and
obstacles. Feature information from digital maps, such as road networks, elev-
ation, and hydrology, can also be incorporated into this rich world model. The
various features are maintained in different layers that are registered together to
provide maximum flexibility in generation of vehicle plans depending on mis-
sion requirements. The world model includes occupancy grids and symbolic
object representations at each level of the hierarchy. Information at different
hierarchical levels has different spatial and temporal resolution. The details of
a world model are as follows:
Low resolution obstacle map and elevation map. The obstacle map consists
of a 2D array of cells [24]. Each cell of the map represents one of the follow-
ing situations: traversable, obstacle (positive and negative), undefined (such as
blind spots), potential hazard, and so forth. In addition, high-level terrain classi-
fication results can also be incorporated in the map (long grass or small bushes,
steps, and slopes). The elevation contains averaged terrain heights.
Mid-resolution terrain feature map. The features used are of two types,
smooth regions and sharp discontinuities [25].
A priori information. This includes multiple resolution satellite maps and
other known information about the terrain.

© 2006 by Taylor & Francis Group, LLC

FRANKL: “dk6033_c001” — 2006/3/31 — 16:42 — page 15 — #15